LSTM not outputting sequence

1 次查看（过去 30 天）

显示更早的评论

John Malik 2019-12-18

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/497298-lstm-not-outputting-sequence

编辑： John Malik 2019-12-26

I am attempting to do sequence-to-sequence classification.

I haveNtime series of

observations each, and each observation collectsp features.

I build a

cell array XTrain. I set XTrain{i} to be the i-th

time series in my database.

I have two classes. I build a

cell array YTrain, where YTrain{i} is a

categorical vector telling me which class is at which time.

Now I build the following network:

inputSize = [p, 1, 1];

filterSize = [2 1];

numFilters = 20;

numHiddenUnits = 128;

numClasses = 2;

layers = [ ...

sequenceInputLayer(inputSize,'Name','input')

sequenceFoldingLayer('Name','fold')

convolution2dLayer(filterSize,numFilters,'Name','conv1')

reluLayer('Name','relu1')

convolution2dLayer(filterSize,numFilters,'Name','conv2')

reluLayer('Name','relu2')

flattenLayer('Name','flatten')

sequenceUnfoldingLayer('Name','unfold')

lstmLayer(numHiddenUnits,'OutputMode','sequence','Name','lstm')

fullyConnectedLayer(numClasses,'Name','fc')

softmaxLayer('Name','softmax')

classificationLayer('Name','classification')];

lgraph = layerGraph(layers);

lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');

maxEpochs = 1;

miniBatchSize = 2;

options = trainingOptions('adam', ...

'ExecutionEnvironment','cpu', ...

'MaxEpochs',maxEpochs, ...

'MiniBatchSize',miniBatchSize, ...

'GradientThreshold',1, ...

'Verbose',false, ...

'Plots','training-progress');

net = trainNetwork(XTrain,YTrain,lgraph,options);

However, if I then run:

YScores = predict(net,XTrain,'MiniBatchSize',1);

the output is a

cell array whose i-th entry is a

vector of class probabilities.

This is INCORRECT. It should be a

vector of class probabilities.

4 个评论
显示 2更早的评论隐藏 2更早的评论

Mohammad Sami 2019-12-26

Can you try putting the sequenceUnfoldingLayer before the flatten layer.

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.sequenceunfoldinglayer.html

John Malik 2019-12-26

编辑：John Malik 2019-12-26

Thanks, I had that first and changed it to this version.

Neither work.

I ended up coding it myself from scratch.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

请先登录，再回答此问题。

类别

AI and Statistics Deep Learning Toolbox Image Data Workflows

在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by